Search

Background Biological drugs have improved the management of immune-mediated inflammatory diseases (IMIDs) despite being associated with important safety issues such as immunogenicity, infections, and malignancies in real-world settings. Objective The aim of this study was to explore the potential of a large Italian multi-database distributed network for use in the postmarketing surveillance of biological drugs, including biosimilars, in patients with IMID. Methods A retrospective cohort study was conducted using 13 Italian regional claims databases during 2010–2019. A tailor-made R-based tool developed for distributed analysis of claims data using a study-specific common data model was customized for this study. We measured the yearly prevalence of biological drug users and the frequency of switches between originator and biosimilars for infliximab, etanercept, and adalimumab separately and stratified them by calendar year and region. We then calculated the cumulative number of users and person-years (PYs) of exposure to individual biological drugs approved for IMIDs. For a number of safety outcomes (e.g., severe acute respiratory syndrome coronavirus 2 [SARS-COV-2] infection), we conducted a sample power calculation to estimate the PYs of exposure required to investigate their association with individual biological drugs approved for IMIDs, considering different strengths of association. Results From a total underlying population of almost 50 million inhabitants from 13 Italian regions, we identified 143,602 (0.3%) biological drug users, with a cumulative exposure of 507,745 PYs during the entire follow-up. The mean age ± standard deviation of biological drug users was 49.3 ± 16.3, with a female-to-male ratio of 1.2. The age-adjusted yearly prevalence of biological drug users increased threefold from 0.7 per 1000 in 2010 to 2.1 per 1000 in 2019. Overall, we identified 40,996 users of biosimilars of tumor necrosis factor (TNF)-α inhibitors (i.e., etanercept, adalimumab, and infliximab) in the years 2015–2019. Of these, 46% (N = 18,845) switched at any time between originator and biosimilars or vice versa. To investigate a moderate association (incidence rate ratio 2) between biological drugs approved for IMIDs and safety events of interest, such as optic neuritis (lowest background incidence rate 10.4/100,000 PYs) or severe infection (highest background incidence rate 4312/100,000 PYs), a total of 43,311 PYs and 104 PYs of exposure to individual biological drugs, respectively, would be required. As such, using this network, of 15 individual biological drugs approved for IMIDs, the association with those adverse events could be investigated for four (27%) and 14 (93%), respectively. Conclusion The VALORE project multi-database network has access to data on more than 140,000 biological drug users (and > 0.5 million PYs) from 13 Italian regions during the years 2010–2019, which will be further expanded with the inclusion of data from other regions and more recent calendar years. Overall, the cumulated amount of person-time of exposure to biological drugs approved for IMIDs provides enough statistical power to investigate weak/moderate associations of almost all individual compounds and the most relevant safety outcomes. Moreover, this network may offer the opportunity to investigate the interchangeability of originator and biosimilars of several TNFα inhibitors in different therapeutic areas in real-world settings. Supplementary Information The online version contains supplementary material available at 10.1007/s40259-021-00498-3.

This is a final study report with the following objectives Primary objectives To monitor and estimate the incidence rates of adverse events of special interest (AESI) in vaccinated and non-vaccinated persons by data source over the period January 1st 2020-October 31st 2021 by brand and dose of vaccine and age of the population. To monitor and estimate the incidence rates of diagnosed COVID-19 in vaccinated and non-vaccinated persons by data source over the period January 1st 2020-October 31st 2021 by brand and dose of vaccine as well as age. To monitor exposure and coverage to COVID-19 vaccines by brand and dose of vaccine as well as age. Secondary objectives To compare the incidence rates of AESIs in the risk window of 28 days after vaccination with dose 1 and/or dose 2 with the incidence rates of AESIs in 2020. To monitor and estimate vaccine exposure, incidence rates of adverse events of special interest. (AESI) and of COVID-19 in vaccinated and non-vaccinated persons over the period January 1st 2020-October 31st 2021 in the at-risk population for developing severe COVID-19 by data source, brand and dose of vaccine as well as age. Methods We used a retrospective cohort design, in 4 electronic health care databases, with periodic updates of the data during the study period (January 2020 -September 2021). Persons entered the cohort on 1/1/2020 and exit upon latest data extraction, death, moving out or the specific events of interest. Person-time after cohort entry was divided in non-exposed person-time, and person-time following vaccination by specific brands, labelled by dose and distance since last vaccination (-1, -2, -3 weeks etc). The source population included all individuals observed in one of the participating data sources for at least one day during the study period (01 January 2020 - last data availability) and who have at least 1 year of data availability before cohort entry, except for individuals with data available since birth. Per event, for calculation of incidence, individuals were followed from cohort entry and contribute to person-time in month (prior to vaccination) and in weeks after vaccination plus specific vaccine exposure (brand & dose) category. Follow-up was censored upon the earliest of date of the event (except for recurrent events), death, exiting the data source, or last data draw-down. For comparison of post-vaccination rates follow-up ended 28 days after each of the vaccine doses, if they had a 2nd dose the intervals post-dose stopped at the date of vaccination with the second. Incidence rates were calculated for 2020 (non-exposed), and after vaccination by vaccine brand, dose and data source. Incidence rates were standardised directly to the Eurostat population and standardised rate differences were calculated using R. Following results of the interim analysis in July 2021, Poisson regression was added to the amended protocol to adjust for measured covariates that were related to the chance of exposure to certain vaccines (age, sex, risk factors for severe covid and prior covid-19). We did not design to adjust for covariates related to the specific outcomes and this study was not designed for causal inference but for monitoring safety. This means that residual confounding may remain, which is why we pre-stated that we classify an association as ‘disproportional’ if the IRR was above 2. Results This study comprised a total of 25,720,158 subjects. We count only the largest population for BIFAP for the total, as the regions with hospital linkage are a subset of the primary care populations. The largest population included was from CPRD with more than 14 million participants. Data locks differed per site: June 30, 2021 in Tuscany, August 31st for BIFAP, August 1st 2021 for PHARMO and May 2021 for CPRD Aurum. At the start of the study 1/1/2020, 34% of the Tuscany population had one or more risk factors for severe COVID-19 disease, and this was around 25% in each of the other data sources (table 2). Median age was highest in Tuscany region (49) and BIFAP-HOSP regions (49). Overall, 12,117,458 persons received a first dose of a Covid-19 vaccine (47.1%) (excluding unknown vaccines manufacturers). Percentage was highest in BIFAP (68.7%). In BIFAP the majority of persons also had received a second dose for each of the vaccine brands. Percentage of full primary regimen of 2-dose primary regimens were lower in other data sources, in particular for AstraZeneca in CPRD, as this vaccine also had the highest distance between dose 1 and 2 in each data source. mRNA vaccines had a short distance between dose 1 and 2 in all sites except for CPRD, where Pfizer also had a mean of 76 days between dose 1 and 2, but only 28 days for Moderna vaccine. In this data instance heterologous schedules were very rare. Vaccination coverage data reflected well the regional/national data for ARS and BIFAP, but were lower for CPRD and PHARMO, probably due to delays in automated feedback on vaccination from immunization registers. We studied the 2020 rates of different AESI. Most AESI were very rare, only the coagulation disorders were more common. We monitored the occurrence of AESI using cumulative weekly rates, and by censoring at 28-day intervals after each dose. The latter was used to compare against background, which was done using age standardized incidence rate differences, and subsequent Poisson analysis adjusting (where possible for age, sex, prior covid-19 and any risk factors for COVID). The table below shows the key results. For most AESI no excess risk was observed following vaccination, 30 event/vaccine/dose combinations showed excess age standardized rate differences and associations in the Poisson analyses (see table below), however after adjustment for factors associated with vaccine roll out, only 10 significant associations of pooled incidence rate ratios remained based on dose 1 and 2 combined. These comprised anaphylaxis after AstraZeneca, TTS after both AstraZeneca and Janssen vaccine, erythema multiforme after Moderna, GBS after Janssen vaccine, SOCV after Janssen vaccine, thrombocytopenia after Janssen and Moderna vaccine and venous thromboembolism after Moderna and Pfizer vaccines. The risk was more than two-fold increased for TTS, SOCV and thrombocytopenia. Conclusions This study has provided many lessons It showed that we could monitor a large number of AESI and COVID-19 across 4 data sources in four countries based on the ConcePTION common data model, and common analytics pipeline, and that semantic harmonization was possible across the different disease terminologies Monitoring could start very early in the vaccination campaign, and repeated updates were possible The same population and data sources were used both to compute background rates, and to retrieve observed events after vaccination. This design avoids a limitation of using, on the one hand, real-world data to assess background rates, and, on the other, spontaneous reporting to assess observed cases: underestimation, if any, is more likely to affect the two periods is a uniform way, thus improving the validity of comparison. Underestimation of an AESI can be discussed, based on the characteristics of the data source in relation with the AESI. For example, ICPC codes do not allow for studying the majority for rare AESI, which affected the ability of PHARMO of monitoring such AESI; or, events that do not require hospitalisation or access to emergency room cannot be studied in the ARS data source. COVID-19 vaccines had very different user patterns across the countries in terms of type, distance between dose 1 and 2 and the populations targeted. We observed strong channelling of the different vaccines that differed across countries AESI incidence rates were mostly very low, especially for neurological, immunological and haematological events. Coagulations disorders and cardiac disorders were more frequent, at the same time such events were those with stronger confounding For several AESI we observed disproportionalities between post-vaccination observed and expected rates. Most of these events had been the topic of regulatory discussions, based on public records such as the haematological events, neurological events and erythema multiforme. In spite of the large numbers of vaccinees, power is limited for the events that are very rare, The research leading to these results was conducted as part of the activities of the EU PE&PV (Pharmacoepidemiology and Pharmacovigilance) Research Network (led by Utrecht University) with collaboration from the Vaccine Monitoring Collaboration for Europe network (VAC4EU). Scientific work for this project was coordinated by the University Medical Center Utrecht. The project has received support from the European Medicines Agency under the Framework service contract nr EMA/2018/28/PE. This document expresses the opinion of the authors, and may not be understood or quoted as being made on behalf of or reflecting the position of the European Medicines Agency or one of its committees or working parties. The authors from BIFAP would like to acknowledge the excellent collaboration of the primary care practitioners and pediatricians, and also the support of the regional authorities participating in the database.

Background: Little is known about the comparative risk of death with atypical or conventional antipsychotics (APs) among persons with cardiovascular or cerebrovascular disease (CCD). Research design and methods: A cohort study was conducted using five Italian claims databases. New atypical AP users with CCD aged ≥65 (reference) were matched to new conventional AP users. Mortality per 100 person-years (PYs) and hazard ratios (HR), estimated using Cox models, were reported. Incidence and risk of death were estimated for persons having drug–drug interactions. Outcome occurrence was evaluated 180 days after AP initiation. Results: Overall 24,711 and 27,051 elderly new conventional and atypical AP users were identified. The mortality rate was 51.3 and 38.5 deaths per 100 PYs for conventional and atypical AP users. Mortality risk was 1.33 (95%CI: 1.27–1.39) for conventional APs. There was no increased mortality risk with single drug–drug interactions (DDIs) vs. no DDI. AP users with ≥1 DDI had a 29% higher mortality risk compared to no DDI in the first 90 days of treatment (HR: 1.29 (95% CI: 1.00–1.67)). Conclusions: Conventional APs had a higher risk of death than atypical APs among elderly persons with CCD. Having ≥1 DDI was associated with an increased risk of death.

In the last years, the DeepMind algorithm AlphaZero has become the state of the art to efficiently tackle perfect information two-player zero-sum games with a win/lose outcome. However, when the win/lose outcome is decided by a final score difference, AlphaZero may play score-suboptimal moves because all winning final positions are equivalent from the win/lose outcome perspective. This can be an issue, for instance when used for teaching, or when trying to understand whether there is a better move. Moreover, there is the theoretical quest for the perfect game. A naive approach would be training an AlphaZero-like agent to predict score differences instead of win/lose outcomes. Since the game of Go is deterministic, this should as well produce an outcome-optimal play. However, it is a folklore belief that "this does not work". In this paper, we first provide empirical evidence for this belief. We then give a theoretical interpretation of this suboptimality in general perfect information two-player zero-sum game where the complexity of a game like Go is replaced by the randomness of the environment. We show that an outcome-optimal policy has a different preference for uncertainty when it is winning or losing. In particular, when in a losing state, an outcome-optimal agent chooses actions leading to a higher score variance. We then posit that when approximation is involved, a deterministic game behaves like a nondeterministic game, where the score variance is modeled by how uncertain the position is. We validate this hypothesis in AlphaZero-like software with a human expert., Comment: Published at 2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA). This version (v2) is a major revision and superseeds version v1

Rationale and background: The global rapid spread of COVID-19 caused by the SARS-CoV-2 triggered the need for developing vaccines to control for this pandemic. This study aimed to generate background incidence rates of adverse events of special interest (AESI) that may be used to monitor benefit-risk profile of COVID-19 vaccines. Data sources: This study included 10 data sources from 7 European countries (Denmark, Germany, France, Italy, Netherlands, Spain, United Kingdom). Data sources contain health insurance data (GePaRD, SNDS), hospitalisation record linkage data (PHARMO, Danish registries (DCE-AU), SIDIAP, ARS) or data from general practitioners (CPRD, PEDIANET, BIFAP, FISABIO). For this final report data from 9 data sources were included. Study size: The study population for the total study comprised approximately 141.6 million individuals. In this final report, a total number of 45 million individuals were included. An update including French data is expected later this year Results This report comprises background rate data on AESI from 6 countries (UK, ES, IT, DK, NL, DE) and 9 data sources(BIFAP, Pedianet (children only), CPRD, ARS, Danish registries, FISABIO, SIDIAP, PHARMO, GeParD). Data from France (SNDS) could not be generated in a timely manner due to administrative constraints in data release. Data sources included different subpopulations based on the availability of numerator data of the observed persontime (Hosp= hospital based, PC= primary care, HOSP-PC= overlap between hospitalization and primary care). This entry also includes the results in excel format and also the links to the codes and event definitions, This protocol has been accepted by EMA as a deliverable of the framework contract No EMA/2018/28/PE. The protocol expresses the expertise of the authors and the ACCESS consortium as well as feedback received from EMA. It may not be understood or quoted as being made on beh behalf, or reflecting the position of the European Medicines Agency or one of its Committees or Working Parties. For questions please contact: m.c.j.sturkenboom@umcutrecht.nl

Enormous progress has been made globally in the use of evidence derived from patients’ clinical information as they access their routine medical care. The value of real-world data lies in their complementary nature compared with data from randomised controlled trials: less detailed information on drug efficacy but longer observational periods and larger, more heterogeneous study populations reflecting clinical practice because individuals are included who would not usually be recruited in trials. Real-world data can be collected in various types of electronic sources, such as electronic health records, claims databases and drug or disease registries. These data sources vary in nature from country to country, according to national healthcare system structures and national policies. In Italy, a growing number of healthcare databases have been used to evaluate post-marketing drug utilisation and safety in the last two decades. The aim of this narrative review is to describe the available Italian sources of real-world data and their contribution to generating post-marketing evidence on drug use and safety. We also discuss the strengths and limitations of the most commonly used Italian healthcare databases in addressing various research questions concerning drug utilisation, comparative effectiveness and safety studies, as well as health technology assessment and other areas.

BACKGROUND: acute myocardial infarction (AMI), ischemic heart diseases (IHDs) and stroke are serious cardiovascular diseases (CVDs) which may lead to hospitalizations, require periodical medical monitoring and life-long drugs use, thus having a high impact on public health and Healthcare Service expenditure. In this contest, Italian Healthcare Administrative Databases (HADs), which routinely collect patient-level information on healthcare services reimbursed by the National Healthcare service, are increasingly used for identification of these CVDs. ObjectiveS: to identify and describe all AMI, IHDs and stroke case-identification algorithms by means of Italian HADs, through the review of papers published in the past 10 years. Methods: this study is part of a project that systematically reviewed case-identification algorithms for 18 acute and chronic conditions by means of HADs in Italy. PubMed was searched for original articles, published between 2007 and 2017, in Italian or English. The search string consisted of a combination of free text and MeSH terms with a common part that focused on HADs and a disease-specific part. All identified papers were screened by two independent reviewers. Pertinent papers were classified according to the objective for which the algorithm had been used, and only articles that used algorithms for primary objectives (I disease occurrence; II population/cohort selection; III outcome identification) were considered for algorithm extraction. The HADs used (hospital discharge records, drug prescriptions, etc.), ICD-9 and ICD-10 codes, ATC classification of drugs, follow-back periods, and age ranges applied by the algorithms have been reported. Further information on specific objective(s), accuracy measures, sensitivity analyses and the contribution of each HAD, have also been recorded. Results: the search strategy has led to the identification of 611 papers for AMI,801 for IHDs and 791 for stroke. Among these,45,12 and 31 papers for AMI, IHDs and stroke respectively, were considered pertinent for inclusion in the systematic review. The majority of the works was published during 2014-2017. The setting of the studies was mainly regional for AMI and stroke, while the majority of IHD’s papers was based on a national multicenter context. By screening full texts, a total of 17,5 and 28 original algorithms for AMI, IHDs and stroke respectively, intended for the above-mentioned objectives, were found. Moreover, 3 original algorithms for STEMI, 3 for NSTEMI, 8 for ischemic stroke and 3 for hemorrhagic stroke were identified. The hospital discharge diagnosis database (HDD) was used in all algorithms. In only a few cases the co-payment exemption registry, drug prescription database, and mortality registry database were used as additional algorithm components. For the same event, there was always a difference of >1 code. External validation was performed in only one case for AMI and stroke identification. Conclusion: a remarkable heterogeneity, in terms of both data sources and codes used, was observed for algorithms aimed to identify AMI, IHDs and stroke in HADs. This was likely due to the paucity of validation studies. Administrative data sources other than HDD remain underutilized.

BACKGROUND: there has been a long-standing, consistent use worldwide of Healthcare Administrative Databases (HADs) for epidemiological purposes, especially to identify acute and chronic health conditions. These databases are able to reflect health-related conditions at a population level through disease-specific case-identification algorithms that combine information coded in multiple HADs. In Italy, in the past 10 years, HAD-based case-identification algorithms have experienced a constant increase, with a significant extension of the spectrum of identifiable diseases. Besides estimating incidence and/or prevalence of diseases, these algorithms have been used to enroll cohorts, monitor quality of care, assess the effect of environmental exposure, and identify health outcomes in analytic studies. Despite the rapid increase in the use of case-identification algorithms, information on their accuracy and misclassification rate is currently unavailable for most conditions. OBJECTIVES: to define a protocol to systematically review algorithms used in Italy in the past 10 years for the identification of several chronic and acute diseases, providing an accessible overview to future users in the Italian and international context. METHODS: PubMed will be searched for original research articles, published between 2007 and 2017, in Italian or English. The search string consists of a combination of free text and MeSH terms with a common part on HADs and a disease-specific part. All identified papers will be screened for eligibility by two independent reviewers. All articles that used/defined an algorithm for the identification of each disease of interest using Italian HADs will be included. Algorithms with exclusive use of death certificates, pathology register, general practitioner or pediatrician data will be excluded. Pertinent papers will be classified according to the objective for which the algorithm was used, and only articles that used algorithms with "primary objectives" (I disease occurrence; II population/cohort selection; III outcome identification) will be considered for algorithm extraction. The HADs used (hospital discharge records, drug prescriptions, etc.), ICD-9 and ICD-10 codes, ATC classification of drugs, follow-back periods, and age ranges applied by the algorithms will be collected. Further information on specific accuracy measures from external validations, sensitivity analyses, and the contribution of each source will be recorded. This protocol will be applied for 16 different systematic reviews concerning eighteen diseases (Hypothyroidism, Hyperthyroidism, Diabetes mellitus, Type 1 diabetes mellitus, Acute myocardial infarction, Ischemic heart disease, Stroke, Hypertension, Heart failure, Congenital heart anomalies, Parkinson's disease, Multiple sclerosis, Epilepsy, Chronic obstructive pulmonary disease, Asthma, Inflammatory bowel disease, Celiac disease, Chronic kidney failure). CONCLUSION: this protocol defines a standardized approach to extensively examine and compare all experiences of case identification algorithms in Italy, on the 18 abovementioned diseases. The methodology proposed may be applied to other systematic reviews concerning diseases not included in this project, as well as other settings, including international ones. Considering the increasing availability of healthcare data, developing standard criteria to describe and update characteristics of published algorithms would be of great use to enhance awareness in the choice of algorithms and provide a greater comparability of results.